def __init__(self,
out_channels,
pretrained=True,
norm=nn.BatchNorm2d,
freeze_bn=False):
super().__init__()
self.freeze_bn = freeze_bn
res = resnet50(pretrained)
res.fc = None
self.inplanes = res.inplanes
self.encoder1 = nn.Sequential(
res.conv1,
res.bn1,
res.relu,
res.maxpool,
res.layer1, # ceil(size / 4)
)
self.ASPP = ASPP(2048, norm=norm)
self.layer4 = res.layer4
self.layer3 = res.layer3
self.encoder2 = nn.Sequential(
res.layer2, # ceil(size / 8)
self.layer3, # ceil(size / 16)
self.layer4,
self.ASPP)
self.layer4[0].dilation = (2, 2)
for m in self.layer4.modules():
if isinstance(m, nn.modules.Conv2d):
m.stride = 1
for m in self.encoder1.modules():
if isinstance(m, nn.modules.ReLU):
m.inplace = True
for m in self.encoder2.modules():
if isinstance(m, nn.modules.ReLU):
m.inplace = True
self.lowDecoder = nn.Sequential(
nn.Conv2d(256, 48, 1, 1, bias=False),
nn.BatchNorm2d(48),
# nn.ReLU(True)
)
self.final = nn.Sequential(nn.Conv2d(304, 256, 3, 1, bias=False),
nn.BatchNorm2d(256), nn.ReLU(True),
nn.Conv2d(256, 256, 3, 1, bias=False),
nn.BatchNorm2d(256), nn.ReLU(True),
nn.Conv2d(256, out_channels, 1))
def __init__(self, num_classes): #use encoder to pass pretrained encoder
super(ERFNet,self).__init__()
self.encoder = Encoder(num_classes)
self.decoder = Decoder(1)
self.aspp = ASPP(dim_in=128,dim_out=128,rate=1,bn_mom=0.007)
self.psp_layer = PyramidPooling('psp', 128, 128,
norm_layer=nn.BatchNorm2d)
def __init__(self):
super(Net, self).__init__()
self.down = True
self.startconv = nn.Conv2d(1, 3, kernel_size=1)
self.basemodel = smp.Unet(
encoder_name='efficientnet-b0',
encoder_weights='imagenet',
in_channels=3,
classes=1
)
self.planes = [32, 48, 136, 384]
self.down = False
self.center = ASPP(self.planes[3], self.planes[2])
def attn_reg(opt, input_size, lossfxn):
img_input = Input(shape=input_size, name='input_scale1')
scale_img_2 = AveragePooling2D(pool_size=(2, 2),
name='input_scale2')(img_input)
scale_img_3 = AveragePooling2D(pool_size=(2, 2),
name='input_scale3')(scale_img_2)
scale_img_4 = AveragePooling2D(pool_size=(2, 2),
name='input_scale4')(scale_img_3)
conv1 = UnetConv2D(img_input, 32, is_batchnorm=True, name='conv1')
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
input2 = Conv2D(64, (3, 3),
padding='same',
activation='relu',
name='conv_scale2')(scale_img_2)
input2 = concatenate([input2, pool1], axis=3)
conv2 = UnetConv2D(input2, 64, is_batchnorm=True, name='conv2')
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
input3 = Conv2D(128, (3, 3),
padding='same',
activation='relu',
name='conv_scale3')(scale_img_3)
input3 = concatenate([input3, pool2], axis=3)
conv3 = UnetConv2D(input3, 128, is_batchnorm=True, name='conv3')
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
input4 = Conv2D(256, (3, 3),
padding='same',
activation='relu',
name='conv_scale4')(scale_img_4)
input4 = concatenate([input4, pool3], axis=3)
conv4 = UnetConv2D(input4, 64, is_batchnorm=True, name='conv4')
pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)
#center = UnetConv2D(pool4, 512, is_batchnorm=True, name='center')
center = ASPP(pool4, 512, name='center')
g1 = UnetGatingSignal(center, is_batchnorm=True, name='g1')
attn1 = AttnGatingBlock(conv4, g1, 128, '_1')
up1 = concatenate([
Conv2DTranspose(32, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
kernel_initializer=kinit)(center), attn1
],
name='up1')
g2 = UnetGatingSignal(up1, is_batchnorm=True, name='g2')
attn2 = AttnGatingBlock(conv3, g2, 64, '_2')
up2 = concatenate([
Conv2DTranspose(64, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
kernel_initializer=kinit)(up1), attn2
],
name='up2')
g3 = UnetGatingSignal(up1, is_batchnorm=True, name='g3')
attn3 = AttnGatingBlock(conv2, g3, 32, '_3')
up3 = concatenate([
Conv2DTranspose(32, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
kernel_initializer=kinit)(up2), attn3
],
name='up3')
up4 = concatenate([
Conv2DTranspose(32, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
kernel_initializer=kinit)(up3), conv1
],
name='up4')
conv6 = UnetConv2D(up1, 256, is_batchnorm=True, name='conv6')
conv7 = UnetConv2D(up2, 128, is_batchnorm=True, name='conv7')
conv8 = UnetConv2D(up3, 64, is_batchnorm=True, name='conv8')
conv9 = UnetConv2D(up4, 32, is_batchnorm=True, name='conv9')
out6 = Conv2D(1, (1, 1), activation='sigmoid', name='pred1')(conv6)
out7 = Conv2D(1, (1, 1), activation='sigmoid', name='pred2')(conv7)
out8 = Conv2D(1, (1, 1), activation='sigmoid', name='pred3')(conv8)
out9 = Conv2D(1, (1, 1), activation='sigmoid', name='final')(conv9)
model = Model(inputs=[img_input], outputs=[out6, out7, out8, out9])
loss = {
'pred1': lossfxn,
'pred2': lossfxn,
'pred3': lossfxn,
'final': losses.tversky_loss
}
loss_weights = {'pred1': 1, 'pred2': 1, 'pred3': 1, 'final': 1}
model.compile(optimizer=opt,
loss=loss,
loss_weights=loss_weights,
metrics=[losses.dsc])
return model
def attn_unet(opt, input_size, lossfxn):
inputs = Input(shape=input_size)
conv1 = UnetConv2D(inputs, 32, is_batchnorm=True, name='conv1')
pool1 = MaxPooling2D(pool_size=(2, 2))(conv1)
conv2 = UnetConv2D(pool1, 32, is_batchnorm=True, name='conv2')
pool2 = MaxPooling2D(pool_size=(2, 2))(conv2)
conv3 = UnetConv2D(pool2, 64, is_batchnorm=True, name='conv3')
#conv3 = Dropout(0.2,name='drop_conv3')(conv3)
pool3 = MaxPooling2D(pool_size=(2, 2))(conv3)
conv4 = UnetConv2D(pool3, 64, is_batchnorm=True, name='conv4')
#conv4 = Dropout(0.2, name='drop_conv4')(conv4)
pool4 = MaxPooling2D(pool_size=(2, 2))(conv4)
#center = UnetConv2D(pool4, 128, is_batchnorm=True, name='center')
center = ASPP(pool4, 128, name='center')
g1 = UnetGatingSignal(center, is_batchnorm=True, name='g1')
attn1 = AttnGatingBlock(conv4, g1, 128, '_1')
up1 = concatenate([
Conv2DTranspose(32, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
kernel_initializer=kinit)(center), attn1
],
name='up1')
g2 = UnetGatingSignal(up1, is_batchnorm=True, name='g2')
attn2 = AttnGatingBlock(conv3, g2, 64, '_2')
up2 = concatenate([
Conv2DTranspose(64, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
kernel_initializer=kinit)(up1), attn2
],
name='up2')
g3 = UnetGatingSignal(up1, is_batchnorm=True, name='g3')
attn3 = AttnGatingBlock(conv2, g3, 32, '_3')
up3 = concatenate([
Conv2DTranspose(32, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
kernel_initializer=kinit)(up2), attn3
],
name='up3')
up4 = concatenate([
Conv2DTranspose(32, (3, 3),
strides=(2, 2),
padding='same',
activation='relu',
kernel_initializer=kinit)(up3), conv1
],
name='up4')
out = Conv2D(1, (1, 1),
activation='sigmoid',
kernel_initializer=kinit,
name='final')(up4)
model = Model(inputs=[inputs], outputs=[out])
model.compile(optimizer=opt,
loss=lossfxn,
metrics=[losses.dsc, losses.tp, losses.tn])
return model
def __init__(self, pretrained=True, num_classes=-1):
super(deeplabv3plus, self).__init__()
self.backbone = None
self.backbone_layers = None
input_channel = 2048
self.aspp = ASPP(dim_in=input_channel,
dim_out=256,
rate=16 // 16,
bn_mom=0.0003)
self.dropout1 = nn.Dropout(0.5)
self.upsample4 = nn.UpsamplingBilinear2d(scale_factor=4)
self.upsample_sub = nn.UpsamplingBilinear2d(scale_factor=16 //
8) #16//4
indim = 728
shallow1_dim = 64
self.shortcut_conv1_1 = nn.Sequential(
nn.Conv2d(indim, shallow1_dim, 1, 1, padding=1 // 2, bias=True),
SynchronizedBatchNorm2d(shallow1_dim, momentum=0.0003),
nn.ReLU(inplace=True),
)
self.cat_conv1_1 = nn.Sequential(
nn.Conv2d(256 + shallow1_dim, 256, 3, 1, padding=1, bias=True),
SynchronizedBatchNorm2d(256, momentum=0.0003),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(256, 256, 3, 1, padding=1, bias=True),
SynchronizedBatchNorm2d(256, momentum=0.0003),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
indim = 256
shallow2_dim = 32
self.shortcut_conv1_2 = nn.Sequential(
nn.Conv2d(indim, shallow2_dim, 1, 1, padding=1 // 2, bias=True),
SynchronizedBatchNorm2d(shallow2_dim, momentum=0.0003),
nn.ReLU(inplace=True),
)
self.cat_conv1_2 = nn.Sequential(
nn.Conv2d(256 + shallow2_dim, 256, 3, 1, padding=1, bias=True),
SynchronizedBatchNorm2d(256, momentum=0.0003),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(256, 256, 3, 1, padding=1, bias=True),
SynchronizedBatchNorm2d(256, momentum=0.0003),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
# self.predict5x5 = nn.Conv2d(256, 256, 5, 1, padding=2)
self.predict5x5 = nn.Sequential(
nn.Conv2d(256, 256, 3, 1, padding=1, bias=True),
SynchronizedBatchNorm2d(256, momentum=0.0003),
nn.ReLU(inplace=True),
nn.Dropout(0.5),
nn.Conv2d(256, 256, 3, 1, padding=1, bias=True),
SynchronizedBatchNorm2d(256, momentum=0.0003),
nn.ReLU(inplace=True),
nn.Dropout(0.1),
)
self.cls_conv = nn.Conv2d(256, num_classes, 1, 1, padding=0)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight,
mode='fan_out',
nonlinearity='relu')
elif isinstance(m, SynchronizedBatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
self.backbone = xception.xception(pretrained=pretrained, os=16)
self.backbone_layers = self.backbone.get_layers()
class DeepLabV3Plus(nn.Module):
def __init__(self,
out_channels,
pretrained=True,
norm=nn.BatchNorm2d,
freeze_bn=False):
super().__init__()
self.freeze_bn = freeze_bn
res = resnet50(pretrained)
res.fc = None
self.inplanes = res.inplanes
self.encoder1 = nn.Sequential(
res.conv1,
res.bn1,
res.relu,
res.maxpool,
res.layer1, # ceil(size / 4)
)
self.ASPP = ASPP(2048, norm=norm)
self.layer4 = res.layer4
self.layer3 = res.layer3
self.encoder2 = nn.Sequential(
res.layer2, # ceil(size / 8)
self.layer3, # ceil(size / 16)
self.layer4,
self.ASPP)
self.layer4[0].dilation = (2, 2)
for m in self.layer4.modules():
if isinstance(m, nn.modules.Conv2d):
m.stride = 1
for m in self.encoder1.modules():
if isinstance(m, nn.modules.ReLU):
m.inplace = True
for m in self.encoder2.modules():
if isinstance(m, nn.modules.ReLU):
m.inplace = True
self.lowDecoder = nn.Sequential(
nn.Conv2d(256, 48, 1, 1, bias=False),
nn.BatchNorm2d(48),
# nn.ReLU(True)
)
self.final = nn.Sequential(nn.Conv2d(304, 256, 3, 1, bias=False),
nn.BatchNorm2d(256), nn.ReLU(True),
nn.Conv2d(256, 256, 3, 1, bias=False),
nn.BatchNorm2d(256), nn.ReLU(True),
nn.Conv2d(256, out_channels, 1))
def forward(self, x):
# Encoder
enc1 = self.encoder1(x)
enc2 = self.encoder2(enc1)
# Decoder
lowfeat = self.lowDecoder(enc1)
dec = F.upsample(enc2, (lowfeat.size(2), lowfeat.size(3)),
mode='bilinear',
align_corners=True)
dec = torch.cat([dec, lowfeat], 1)
final = self.final(dec)
final = F.upsample(final, (x.size(2), x.size(3)),
mode='bilinear',
align_corners=True)
final = F.log_softmax(final, dim=1)
return final
def train(self, mode=True):
super(DeepLabV3Plus, self).train(mode)
if self.freeze_bn:
for module in self.modules():
if isinstance(module, nn.modules.BatchNorm1d):
module.eval()
if isinstance(module, nn.modules.BatchNorm2d):
module.eval()
if isinstance(module, nn.modules.BatchNorm3d):
module.eval()
def changeStride(self, ostride=8):
self.layer4[0].dilation = (4, 4)
for m in self.layer4.modules():
if isinstance(m, nn.modules.Conv2d):
m.stride = 1
self.layer3[0].dilation = (2, 2)
for m in self.layer3.modules():
if isinstance(m, nn.modules.Conv2d):
m.stride = 1
self.ASPP.changeStride()